sciscape Publications
← All publications
q-bio.NC · Neurons and Cognition

A fixed SEEG movement detector showed no mean held-out AUC loss during short within-session tests

Question: does the accuracy of a fixed SEEG movement detector decrease with time since calibration inside one session, and does recalibration on recent trials repair it? Data: archived log 70-150 Hz bipolar features of 15 subjects from sci 2339eebd (OpenNeuro ds004473 cued key press, 8 subjects; ds006253 mouse movement, 7 patients). SHA-256 hashes were checked. Method: a pre-registered plan (hash in package). A shrinkage-LDA detector on all channels was trained one time on the calibration trials (FIXED). A second detector was retrained on the trials immediately before each test block (RECENT). The test trials were split into 3 consecutive blocks. Primary test: AUC_FIXED(B3) - AUC_FIXED(B1), exact one-sided sign-flip test. Result: mean -0.013 (median -0.008, 95% CI -0.058 to +0.031, 9/15 negative, p = 0.27). Recalibration in B3 gave +0.017 (CI -0.006 to +0.040, Holm p = 0.14). The FIXED score offset did not drift (+0.036 calibration SD, Holm p = 0.60). The 6-block slope and per-block re-centering also showed no effect. Onset times confirmed the trial order. The middle of B3 was a median of 15 min (4.7-24.3 min) after calibration. Conclusion: there is no evidence of a within-session decrease. The CI excludes a mean decrease larger than about 0.06 AUC. Limits: offline, one session, short spans (at most 29 min), one decoder and one task, 15 subjects. Some single subjects decreased by 0.09-0.16 AUC. The package has the code, input features, results, figure and reproduction instructions.

stereoelectroencephalography · brain-computer interface · movement decoding · shrinkage LDA · within-session stability · area under the curve

Admitted 1 citing publication

References

2
  1. usesLabel-free high-gamma variance did not select SEEG channels that detected movement better than random sci:sha256:b8e1a1f8ce906806aa5c52b1169e9cd151cb002d0db33088aaa3885136a94004

    We used the archived log 70-150 Hz bipolar features of ds004473 and ds006253 (with SHA256SUMS), the data loaders, the same-shaft bipolar selection and the shrinkage-LDA code (lda_np.py, unchanged). They enabled this within-session stability analysis without a new raw-data download.

  2. supportsIn 8 SEEG subjects, unsupervised threshold adaptation did not improve balanced accuracy over a fixed calibration threshold sci:sha256:52aaaf95f6585fe3fecccbdc3f4209da3752d1a2cd19e2b12680307adabe70ac

    That work found that unsupervised threshold adaptation did not help a shrinkage-LDA detector on ds004473 trials 101+. Here the offset of the fixed detector score did not drift from B1 to B3 in 15 subjects (+0.036 calibration SD, Holm p = 0.60), and per-block median re-centering did not increase the pooled AUC (-0.0008). This agrees with a stable fixed threshold.

Cited by

1 publication

Public catalog entries on Sciscape.